• Research Question
• Project Background
• Significance of Research
• Data
• Methods
• Selected Findings
• Known Issues
• Future Research

Can the sentiment expressed in trail users' tweets help to assess the effectiveness of Environmental Stewardship Scheme (ESS) Agreements? A case study of the Pennine Way National Trail.

• England has 15 National Trails which pass through diverse landscapes and expanses of agricultural land
• Original concept based upon the Appalachian Trail in the eastern US
• April 25th 2015 marks the 50th anniversary of England’s first National Trail - The Pennine Way (PWNT)
• Annually 12 million visits made to England’s National Trails (Ramblers, 2012).
• 300,000 average annual visitors to the PWNT (based upon Natural England trail counter data)

• The Pennine Way National Trail travels 431 km (268 miles) along the central upland spine of England between Edale, Derybyshire to Kirk Yelthom in the Scottish Borders

• Provides government-financed payments to farmers or land managers in return for an environmentally sensitive approach to farming
• Most widespread approach to environmental management in England with agreements in place on over 70% of agricultural land
• 74.08% within 5km of Pennine Way
• Farmers can select from 65 management options which contribute toward a per-hectare total
• Meet point thresholds to secure a bi-annual payment.

• National Trails aim to provide rewarding natural adventures
  
• Quality of the scenery and landscape previously identified as primary attraction (The Countryside Agency, 2005; Wood-Gee, 2008).
• 2015 marks the 50 year anniversary of England’s first National Trail

• Measurement of ESS effectiveness has evolved
• Traditionally focused on participation levels
• Now more focused on environmental benefits delivered

• Currently no efficient method to obtain trail user opinions of ESS
• Does ESS help maintain landscape quality and character?
• Is social media data a cost-effective and efficient source?

• 60,466 geotagged tweets (generally assumed to be 1-3% of all tweets)
• Collected between 2014-06-03 and 2014-07-25 (actually 49 of the 52 days) (Lovelace, 2014)

GPX file of PWNT route used to generate the 5km and 25km trail buffers (corridors)

• Shapefile of ESS agreements
• 1717 ESS agreements in place within 5km of PWNT;
• 6773 within 25km (25km used for viewshed analyses)

• For viewshed analysis
• Shuttle Radar Thematic Mapper (SRTM) data (Pope, 2009)
• 90m resolution DEM

• The 2007 Land Cover Map (Morton et al., 2011)
• Raster 25m resolution

• Spatial selection - Twitter dataset was reduced to only the tweets which originated from within 5km of the PWNT
• Lexical Selection - relevant tweets selected using NLP; 161 remaining
• Data cleaning - The text of each tweet was processed and ‘cleaned' of spurious characters and Emoji replaced with text equivalent
• Sentiment Analysis (SentiStrength)
• Viewshed Analysis (of each overall positive & overall negative tweet)

SentiStrength was used as the sentiment analysis tool (Thelwall et al., 2010) SentiStrength Homepage

• Specifically designed for data from the social web
• Can analyse both polarity and strength of sentiment within text – although not included here
• The sum of the positive sentiment score and negative sentiment score provided the overall sentiment score for each tweet.

Viewshed analyses were conducted for each overall positive and overall negative tweet: - Viewshed Area - Land Cover - Ruggedness (St. dev. of elevation) - Extent of ESS agreements

Link to interactive web map of PWNT user sentiment

• 40 overall positive tweets; 7 from within ESS
• 13 overall negative tweets; 2 from within ESS
• 108 overall neutral (non-sentiment bearing) tweets; 25 from within ESS

Conclusion - Sample too small to assess the effectiveness of ESS from the trail user perspective.

• Findings not conclusive but a process of extracting trail user sentiment has been proposed
• Need to collect data over longer period of time
• Use hashtag and run awareness campaign?

• Twitter API accessibility
• Data representation of social media data
• Danger of digital divide
• Mobile phone coverage not considered; PWNT is remote, spatial bias
• No consideration of the specific ESS options adopted by farmers and landowners.

• 108 tweets are not sentiment-bearing – classed as neutral
• 94 (87.03%) contained a URL within the TweetText
• Raw data - 16.18% contained a URL
• Sharing URLs has been identified as a significant aspect of Twitter use (boyd et al., 2010)
• Neutral tweets are objective with sentiment-bearing image attached?
• What sentiment is missing from these tweets?
• What do the images ‘say’?